Narada vs tl;dv — EU data residency, or no data that needs residency

If your objection to cloud notetakers is procurement — a DPA, a residency clause, a SOC 2 report — tl;dv is the better answer and this page will not talk you out of it. It hosts in the EEA, names its sub-processors publicly, and publishes a security page you can hand to a reviewer. If your objection is the upload itself, no paperwork fixes that, and Narada takes the other route: recording and transcription happen on your Mac, so there is nothing in a data centre to write a clause about.

That is the whole comparison. The detail still matters, because “bot-free” and “local” are two different words that a lot of pages treat as one.

Who each one is for

tl;dv is a team tool. Recordings land in a shared, searchable library; seats are billed per person; a bot joins Zoom, Google Meet and Teams as a named participant with an avatar, so nobody has to be told the call is being recorded. The free tier is real rather than a trial, and the compliance surface is what a European security review expects: GDPR, SOC 2 and EU AI Act badges, EU hosting, privately hosted AI on request at Enterprise.

Narada is a single-user Mac app. It taps your microphone and your Mac’s output, transcribes on the machine, and writes an .m4a plus plain text into a folder you picked. Nothing joins the call. No workspace, no sharing, no admin console, no Windows build, no phone app. If a teammate needs the notes, you paste them somewhere yourself.

The difference that actually matters

tl;dv’s desktop app is bot-free, and that is genuinely useful — no third party in the attendee list, and their blog says it “captures audio directly from your computer on any platform, not just the big three”. But bot-free is a statement about the meeting, not about the data. Same blog post:

Both modes store recordings in the same central, searchable library with a full audit trail.

So the audio still leaves your Mac, and transcription and summarisation still happen on tl;dv’s infrastructure. Nothing on any tl;dv page claims on-device transcription. That is not a gotcha — a central library is the point of a team product.

Narada’s transcription runs through Core ML on the Mac using Parakeet TDT v3, so the audio file stays put. The honest asterisk: turn on Narada’s cloud AI (Pro AI, or your own OpenAI key) and the transcript text goes to OpenAI after an explicit consent step. Audio never does. Use On-device AI instead — Apple Silicon only, one-time 3 GB or 7 GB model download — and nothing leaves after that download, at the cost of lighter notes than the cloud writes.

Side by side

Where tl;dv is the better choice

If a security questionnaire stands between you and the tool, pick tl;dv. Narada has nothing to put in the boxes: no SOC 2 report, no DPA to countersign, no sub-processor list, no residency commitment, no vendor-risk pack. It is an indie Mac app, and “the data never leaves the employee’s laptop” is an architecture, not an attestation. Some reviewers accept that reasoning. Many require a document, and they are not being unreasonable.

tl;dv gives them documents. The privacy policy commits to EEA processing, names the hosts (Google Cloud, Wasabi and Hetzner data centres “including in Germany and Finland”), and names Anthropic and Google Vertex among its sub-processors rather than listing vague categories. The AI training clause is unusually specific:

Tldx Solutions GmbH does not use Customer Content, including meeting recordings, transcripts, notes, files, or other data processed through the Services, to train, fine-tune, or improve foundation models, large language models, or other generative AI models for the benefit of tldx Solutions GmbH or any third party.

Their security page goes further than most on what the LLM sees: metadata is anonymised, and “We are chunking your meetings into small pieces and randomizing the sequence order with Anthropic. Anthropic will never be able to access more than a short sequence of your meeting at once.”

Also pick tl;dv if your team is on Windows or mixed, if you want everyone’s meetings in one library with shared search, if you want the bot’s visible presence to handle disclosure, or if you need a language Parakeet does not cover — tl;dv claims 30+, and Narada’s 25 are all European, with no Chinese, Japanese, Korean, Arabic, Hindi or Turkish.

Where Narada is the better choice

Pick Narada if the upload is the problem rather than the paperwork. Client calls under NDA, a legal or therapy practice, an internal conversation about a reorganisation, a candidate interview — cases where the honest answer to “who else has a copy of this recording?” has to be “nobody”. No residency setting produces that answer, because residency describes which data centre holds the file, not whether one does.

Pick it if you are one person and a per-seat subscription to a library you will never share is the wrong shape of product. Pick it if you want the files: one folder per meeting holding the audio, the merged transcript, the raw mic and system transcripts, a Markdown summary and a JSON analysis. Plain text you can grep, back up, or feed into something else. There is no export step because there is nothing to export from.

And pick it if the 3-hour cap or the 40-recordings-a-week ceiling would bite. tl;dv’s 3-hour limit applies on every tier, Enterprise included. Narada records as long as the disk holds out, and keeps recording a meeting that runs past its scheduled end instead of cutting at the calendar boundary.

One more thing neither product’s marketing mentions: Narada checks each finished recording against itself — frames submitted versus frames actually in the file — and shows an “Incomplete” warning when audio was lost. Audio can be lost, on any recorder, and a silently short file is worse than a flagged one. Narada does not recover the missing audio. It tells you.

Privacy and where your audio goes

tl;dv, on hosting, from their privacy policy:

All our services are hosted in Europe (within the European Economic Area) with the exception of part of our service involving artificial intelligence, depending on your choice of hosting location. tldx may use large language models provided by Anthropic via Google Cloud Vertex AI to generate written summaries or other derived content.

The same page says “limited portions of meeting transcripts may be processed either: within the European Union … or within the United States of America depending on the AI hosting location that you select in your account’s preferences”. Their security page adds “our data is stored encrypted using AES-256 algorithm”, “tl;dv is SOC2 compliant” and “tl;dv is GDPR-compliant for all users”.

Narada, as the app itself puts it: raw audio never leaves the Mac in any mode, and transcription and speaker separation are on-device. In Recorder-only and On-device AI modes nothing is sent for AI at all. In Pro AI the transcript text goes through Narada’s proxy to OpenAI along with a per-install device identifier and a billing identifier; with your own key it goes to OpenAI directly. Both need explicit consent, and revoking it drops the app back to no AI. On training, the app says: “Narada uses the OpenAI API (not ChatGPT). OpenAI does not use API inputs to train its models by default” — a statement about OpenAI’s policy, not a promise Narada can make for them.

Your files sit unencrypted in your folder, deliberately, so other tools can read them. That makes disk encryption your job — turn on FileVault.

Neither app makes a recording lawful. tl;dv says as much about their own bot-free mode: “The disclosure responsibility sits with the user in bot-free mode … anyone using it should still let participants know the conversation is being captured.” Say it out loud at the start of the call.

Pricing, plainly

Narada. Free forever as a complete recorder: calendar-aware recording, mic plus system audio, on-device transcription with speaker separation, labels, notes, search, storage controls, the Claude Desktop connector. Pro AI is $6.99 a month or $69.99 a year with a 3-day trial, and buys cloud summaries, the follow-ups ledger, chat over your archive and pre-meeting briefs — free instead in On-device AI mode, on an Apple Silicon Mac with the RAM for it.

tl;dv. Per seat, subscription only, and the number depends where you are: the pricing page and tl;dv’s own billing API both answer in the currency of the requesting IP. Checked from a Ukrainian IP on 4 August 2026 it showed Free at ₴0 (“No credit card required”), Pro at ₴832 “per seat/month, billed annually”, Business at ₴1,199 on the same basis, Enterprise as “Contact us for demo and pricing”, plus “Annually UP TO 40% OFF” and a purchasing-parity coupon. No dollar figure appears here on purpose — tl;dv’s own blog posts disagree about the Business price and the page would not render in dollars, so check it from your own connection.

The free tier is generous about recording and strict about AI. Verbatim from the Free column: “Unlimited video recordings”, “Unlimited transcription in 30+ languages”, then “AI features usage limits: 10x meetings with AI notes included / 10x Ask AI queries included”. The comparison table adds what decides it: retention “up to 3 months”, 40 web recordings a week, 3 hours per recording, 2 simultaneous meetings, uploads up to 5. A trial of the paid features exists; tl;dv does not publish its length.

So tl;dv Free is a good way to record and transcribe meetings, three months at a time. Past ten AI-noted meetings it is a paid product.

How to try both

Same week, same kind of call, then compare the transcripts rather than the marketing.

1. Install tl;dv’s desktop app and record one call with the bot and one without. Notice what the other participants say about the bot — that reaction is data.
2. Install Narada from the Mac App Store, grant Microphone and System Audio Recording (it never asks for Screen Recording), point it at a folder and record a comparable call. The first transcription downloads the speech model, roughly 600 MB.
3. Read both transcripts. Narada labels your lines with your name, because your microphone is a separate track, and the other side as Speaker 1, Speaker 2 — no names, because it has no way to know them.
4. Then answer the question this page is about: do you need a document saying where the data is, or do you need there to be no data?

Document: tl;dv. No data, on a Mac, on your own: Narada. For the recording mechanics, the Zoom recording guide explains why system audio needs its own macOS permission.

Sources

• tl;dv, Pricing — https://tldv.io/app/pricing/ — tiers, free-tier bullets, comparison table, FAQ, badges. Rendered from a Ukrainian IP, so prices displayed in UAH.
• tl;dv, Privacy Policy — https://tldv.io/privacy/ — EEA hosting, the AI carve-out, retention, the AI-training clause, sub-processors.
• tl;dv, Security commitment — https://tldv.io/features/security-commitment/ — AES-256, SOC2, GDPR, the Anthropic anonymisation and chunking description.
• tl;dv, Desktop app — https://tldv.io/desktop-app/ — system-audio recording with no bot; macOS M-series, macOS Intel and Windows downloads.
• tl;dv, Is bot-free recording legal? — https://tldv.io/blog/is-bot-free-recording-legal/ — the default visible bot, one shared library for both modes, disclosure responsibility.

All five checked on 4 August 2026. Narada facts are from the shipping build 1.2 (macOS 15.6+) and its own settings and consent text, verified against source the same day.